TSMC, the world’s largest semiconductor foundry, reported its fourth consecutive quarter of record profit on April 16, with net income rising 58% year-over-year to NT$572.48 billion ($17.6 billion). Revenue hit NT$1.134 trillion ($35 billion), beating LSEG SmartEstimate consensus of NT$1.127 trillion. CEO C.C. Wei used the earnings call to name a specific architectural shift as the force behind the numbers: the transition “from generative AI and the query mode to agentic AI and command and action mode.”

The Quote That Matters

Wei’s framing on the call was unusually precise for a semiconductor earnings release. “AI-related demand continues to be extremely robust,” he told analysts, according to CNBC. He then attributed the demand acceleration to the shift from query-based AI (a user asks, a model responds) to agentic AI (an agent plans and executes multi-step tasks autonomously). “This is driving the need for more and more computation, which supports the robust demand for leading-edge silicon,” Wei said, as reported by Manufacturing Dive.

That distinction matters because it puts a specific label on what’s consuming the chips. Agentic workloads require persistent compute, tool calls, multi-step reasoning loops, and parallel execution, all of which burn more tokens per task than a single-shot query. When the CEO of the company that fabricates every major AI chip (NVIDIA H200/GB200, Apple Silicon, AMD Instinct) names that pattern as the demand driver, it is the most credible market signal available.

The Numbers

TSMC raised its full-year 2026 revenue guidance from “close to 30%” to “above 30%” growth in U.S. dollar terms. Q2 revenue is projected at $39 billion to $40.2 billion, a 10% sequential increase, according to CNBC.

The high-performance computing division, which includes AI and 5G applications, accounted for 61% of first-quarter revenue. Advanced chips (7nm or smaller) made up 74% of total wafer revenue. Sub-3nm shipments, the most compute-dense nodes, represented 25%.

William Li, senior analyst at Counterpoint Research, told CNBC that demand has pushed manufacturing capacity to its limits. “The narrative for 2026 is as much about resource constraints as it is about growth,” Li said. “Demand still significantly outpaces supply and isn’t showing any major sign of slowing down.”

Capital expenditure will land at the high end of the previously guided $52 billion to $56 billion range, reflecting expansion across Taiwan, Arizona, and Japan. Construction on TSMC’s third Arizona fab has begun, and permits for a fourth are in progress, as part of the company’s total $165 billion U.S. investment commitment, according to Manufacturing Dive.

The Supply Chain Signal

TSMC’s earnings call comes during a week when four major desktop AI agent systems launched (OpenAI Codex computer use, Google Gemini for Mac, Claude Cowork GA, and Perplexity Personal Computer), SoftBank announced a commercial AI-native smartphone OS, and multiple enterprise platforms shipped agentic capabilities. Wei’s quote connects those product launches to the hardware layer: every agent loop, every tool call, every multi-step task flows through TSMC silicon. The 58% profit surge and the “sold-out environment” that Counterpoint describes are the financial translation of the agentic AI buildout happening across the software stack.

TSMC shares trade on the New York Stock Exchange under the ticker TSM.